Clusternet is a CNCF Sandbox add-on, licensed Apache-2.0 and written in Go, that lets platform teams manage and visit Kubernetes clusters running on public cloud, private cloud, hybrid cloud or the edge as though those clusters were running locally.

What it is

Clusternet, short for Cluster Internet, is an open-source add-on for Kubernetes that treats a fleet of clusters as a single manageable surface. It sets up network tunnels in a configurable way, then exposes a Kubernetes-styled API so an operator can reach a managed cluster or a Kubernetes service with an ordinary KubeConfig. The project lives squarely in the cloud-native ecosystem and is listed as a CNCF Sandbox project. It supports clusters on AWS, Google Cloud, Tencent Cloud and Alibaba Cloud, on-premise clusters, any Certified Kubernetes Distribution such as k3s, and clusters running at the edge, across linux/amd64, linux/arm64, linux/ppc64le, linux/s390x, linux/386 and linux/arm.

The concrete problem it solves is tool sprawl. Teams running workloads in several places tend to accumulate a separate management tool per cluster or per provider, and Clusternet removes that need by making every cluster reachable through one interface. Beyond access, it deploys and coordinates applications to multiple clusters from a single set of APIs hosted in a hosting cluster, so the same control point that opens a tunnel also drives scheduling and rollout.

Key capabilities

• Multi-cluster management and governance for clouds, on-premise and edge, covering Kubernetes above v1.17.x with documented version skew handling.
• Automatic discovery and registration of clusters created by cluster-api.
• Visiting managed clusters under dynamic RBAC rules, rather than a fixed set of credentials.
• Cluster auto-labelling driven by Node Feature Discovery.
• A scheduling framework with in-tree and out-of-tree plugins, offering replication scheduling, static dividing scheduling by weight, dynamic dividing scheduling by capacity, and subgroup cluster scheduling.
• A cluster resource predictor framework with in-tree and out-of-tree implementations and several deployment topologies.
• Resource handling for Kubernetes native objects such as Deployment and StatefulSet, for CRDs, and for Helm charts including OCI-based Helm charts, with resource interpretations handled by in-tree or out-of-tree controllers.

Who uses it and how

• Platform teams running fleets across several public clouds, who need one scheduling and deployment path instead of per-provider tooling.
• Edge operators, where clusters sit in locations that a central team cannot reach directly and tunnels do the work.
• Teams already provisioning with cluster-api, whose new clusters register themselves into Clusternet automatically.
• Hosting-cluster operators who register the parent cluster as a child of itself so that its own workloads are scheduled through the same mechanism as every other cluster.
• Teams standardising on k3s or another Certified Kubernetes Distribution, who want a uniform management layer above heterogeneous distributions.

Getting started

The README excerpt supplied here contains no quickstart, install command, image name or manifest path, so its install route cannot be quoted from these facts. The project points readers to clusternet.io, and the Go module is published at github.com/clusternet/clusternet.

How it compares

No list of paid products that Clusternet replaces is provided in the facts, and no directly comparable alternative tool is named either. The only related projects named are ones Clusternet manages or integrates with, such as k3s and cluster-api, so it stands alone in this registry.

When to use it — and when not to

A self-hoster must run and operate a hosting cluster with its control-plane components, and must keep every managed cluster within the supported Kubernetes version range above v1.17.x. It is not the right pick for a single-cluster deployment, where the added control plane buys nothing, and not for teams unwilling to run their own management plane. The README excerpt available here is thin on installation detail, and the repository carries 72 open issues.